The Role of Defensible Space for Residential Structure Protection During Wildfires

Abstract

With the potential for worsening fire conditions, discussion is escalating over how to best reduce effects on urban communities. A widely supported strategy is the creation of defensible space immediately surrounding homes and other structures. Although state and local governments publish specific guidelines and requirements, there is little empirical evidence to suggest how much vegetation modification is needed to provide significant benefits. We analysed the role of defensible space by mapping and measuring a suite of variables on modern pre-fire aerial photography for 1000 destroyed and 1000 surviving structures for all fires where homes burned from 2001 to 2010 in San Diego County, CA, USA. Structures were more likely to survive a fire with defensible space immediately adjacent to them. The most effective treatment distance varied between 5 and 20 m (16–58 ft) from the structure, but distances larger than 30 m (100 ft) did not provide additional protection, even for structures located on steep slopes. The most effective actions were reducing woody cover up to 40% immediately adjacent to structures and ensuring that vegetation does not overhang or touch the structure. Multiple-regression models showed landscape-scale factors, including low housing density and distances to major roads, were more important in explaining structure destruction. The best long-term solution will involve a suite of prevention measures that include defensible space as well as building design approach, community education and proactive land use planning that limits exposure to fire.

Full text

The role of defensible space for residential structure
protection during wildfires
Alexandra D. Syphard
A
,
D
,Teresa J. Brennan
B
and Jon E. Keeley
B
,
C
A
Conservation Biology Institute, 10423 Sierra Vista Avenue, La Mesa, CA 91941, USA.
B
US Geological Survey Western Ecological Research Center, Three Rivers, CA 93271, USA.
C
Department of Ecology & Evolutionary Biology, University of California, 612 Charles E. Young
Drive, South Los Angeles, CA 90095-7246, USA.
D
Corresponding author. Email: asyphard@consbio.org
Abstract. With the potential for worsening fire conditions, discussion is escalating over how to best reduce effects on
urban communities. A widely supported strategy is the creation of defensible space immediately surrounding homes
and other structures. Although state and local governments publish specific guidelines and requirements, there is little
empirical evidence to suggest how much vegetation modification is needed to provide significant benefits. We analysed
the role of defensible space by mapping and measuring a suite of variables on modern pre-fire aerial photography for 1000
destroyed and 1000 surviving structures for all fires where homes burned from 2001 to 2010 in San Diego County, CA,
USA. Structures were more likely to survive a fire with defensible space immediately adjacent to them. The most effective
treatment distance varied between 5 and 20 m (16–58 ft) from the structure, but distances larger than 30 m (100 ft) did not
provide additional protection, even for structures located on steep slopes. The most effective actions were reducing woody
cover up to 40% immediately adjacent to structures and ensuring that vegetation does not overhang or touch the structure.
Multiple-regression models showed landscape-scale factors, including low housing density and distances to major roads,
were more important in explaining structure destruction. The best long-term solution will involve a suite of prevention
measures that include defensible space as well as building design approach, community education and proactive land use
planning that limits exposure to fire.
Received 16 September 2013, accepted 30 May 2014, published online 14 October 2014
Introduction
Across the globe and over recent decades, homes have been
destroyed in wildfires at an unprecedented rate. In the last
decade, large wildfires across Australia, southern Europe,
Russia, the US and Canada have resulted in tens of thousands of
properties destroyed, in addition to lost lives and enormous
social, economic and ecological effects (Filmon 2004;Boschetti
et al. 2008;Keeley et al. 2009;Blanchi et al. 2010;Vasquez
2011). The potential for climate change to worsen fire condi-
tions (Hessl 2011), and the projection of continued housing
growth in fire-prone wildlands (Gude et al. 2008) suggest that
many more communities will face the threat of catastrophic
wildfire in the future.
Concern over increasing fire threat has escalated discussion
over how to best prepare for wildfires and reduce their effects.
Although ideas such as greater focus on fire hazard in land use
planning, using fire-resistant building materials and reducing
human-caused ignitions (e.g. Cary et al. 2009;Quarles et al.
2010;Syphard et al. 2012) are gaining traction, the traditional
strategy of fuels management continues to receive the most
attention. Fuels management in the form of prescribed fires or
mechanical treatments has historically occurred in remote,
wildland locations (Schoennagel et al. 2009), but recent studies
suggest that treatments located closer to homes and communi-
ties may provide greater protection (Witter and Taylor 2005;
Stockmann et al. 2010;Gibbons et al. 2012). In fact, one of the
most commonly recommended strategies in terms of fuels and
fire protection is to create defensible space immediately around
structures (Cohen 2000;Winter et al. 2009). Defensible space is
an area around a structure where vegetation has been modified,
or ‘cleared,’ to increase the chance of the structure surviving a
wildfire. The idea is to mitigate home loss by minimising direct
contact with fire, reducing radiative heating, lowering the
probability of ignitions from embers and providing a safer place
for fire fighters to defend a structure against fire (Gill and
Stephens 2009;Cheney et al. 2001). Many jurisdictions provide
specific guidelines and practices for creating defensible space,
including minimum distances that are required among trees and
shrubs as well as minimum total distances from the structure.
These distances may be enforced through local ordinances or
state-wide laws. In California, for example, a state law in
2005 increased the required total distance from 9 m (30 ft) to
30 m (100 ft).
Despite these specific guidelines on how to create defensible
space, there is little scientific evidence to support the amount
and location of vegetation modification that is actually effective
CSIRO PUBLISHING
International Journal of Wildland Fire
http://dx.doi.org/10.1071/WF13158
Journal compilation ÓIAWF 2014 www.publish.csiro.au/journals/ijwf

at providing significant benefits. Most spacing guidelines and
laws are based on ‘expert opinion’ or recommendations from
older publications that lack scientific reference or rationale
(e.g. Maire 1979;Smith and Adams 1991;Gilmer 1994).
However, one study has provided scientific support for, and
forms the basis of, most guidelines, policy and laws requiring a
minimum of 30 m (100 ft) of defensible space (Cohen 1999,
2000). The modelling and experimental research in that study
showed that flames from forest fires located 10–40 m (33–131 ft)
away would not scorch or ignite a wooden home; and case studies
showed 90% of homes with non-flammable roofs and vegetation
clearance of 10–20 m (33–66 ft) could survive wildfires (Cohen
2000). However, the models and experimental research in that
study focussed on crown fires in spruce or jack pine forests, and
the primary material of home construction was wood. Therefore,
it is unknown how well this guideline applies to regions domi-
nated by other forest types, grasslands, or nonforested woody
shrublands and in regions where wooden houses are not the norm.
Some older case studies showed that most homes with non-
flammable roofs and 10–18 m (33–ft) of defensible space
survived the 1961 Bel Air fire in California (Howard et al.
1973); most homes with non-flammable roofs and more than
10 m (33 ft) of defensible space also survived the 1990 Painted
Cave fire (Foote and Gilless 1996). Also, several fire-behaviour
modelling studies have been conducted in chaparral shrublands.
One study showed that reducing vegetative cover to 50% at
9–30 m (30–ft) from structures effectively reduced fireline inten-
sity and flame lengths, and that removal of 80% cover would
result in unintended consequences such as exotic grass invasion,
loss of habitat and increase in highly flammable flashy fuels
(A. Fege and D. Pumphrey, unpubl. data). Another showed that
separation distances adequate to protect firefighters varied
according to fuel model and that wind speeds greater than
23 km h
1
negated the effect of slope, and wind speed above
48 km h
1
negated any protective effect of defensible space
(F. Bilz, E. McCormick and R. Unkovich, unpubl. data, 2009).
Results obtained through modelling equations of thermal radia-
tion also found safety distances to vary as a function of fuel type,
type of fire, home construction material and protective garments
worn by firefighters (Za´rate et al. 2008).
Although there is no empirical evidence to support the need
for more than 30 m (100 ft) of defensible space, there has been a
concerted effort in some areas to increase this distance, particu-
larly on steep slopes. In California, a senate bill was introduced
in 2008 (SB 1618) to encourage property owners to clear 91 m
(300 ft) through the reduction of environmental regulations and
permitting needed at that distance. Although this bill was
defeated in committee, many local ordinances do require home-
owners to clear 91 m (300 ft) or more, and there are reports that
some people are unable to get fire insurance without 91 m
(300 ft) of defensible space (F. Sproul, pers. comm.). In contrast,
homeowner acceptance of and compliance with defensible
space policies can be challenging (Winter et al. 2009;Absher
and Vaske 2011), and in many cases homeowners do not create
any defensible space.
It is critically important to develop empirical research that
quantifies the amount, location and distance of defensible space
that provides significant fire protection benefits so that guide-
lines and policies are developed with scientific support.
Data that are directly applicable to southern California are
especially important, as this region experiences the highest
annual rate of wildfire-destroyed homes in the US. Not having
sufficient defensible space is obviously undesirable because of
the hazard to homeowners. However, there are clear trade-offs
involved when vegetation reduction is excessive, as it results in
the loss of native habitats, potential for increased erosion and
invasive species establishment, and it potentially even increases
fire risk because of the high flammability of weedy grasslands
(Spittler 1995;Keeley et al. 2005;Syphard et al. 2006).
It is also important to understand the role of defensible space
in residential structure protection relative to other factors that
explain why some homes are destroyed in fires and some are not.
Recent research shows that landscape-scale factors, such as
housing arrangement and location, as well as biophysical vari-
ables characterising properties and neighbourhoods such as
slope and fuel type, were important in explaining which homes
burned in two southern California study areas (Syphard et al.
2012;2013). Understanding the relative importance of different
variables at different scales may help to identify which combi-
nations of factors are most critical to consider for fire safety.
Our objective was to provide an empirical analysis of the role
of defensible space in protecting structures during wildfires in
southern California shrublands. Using recent pre-fire aerial
photography, we mapped and measured a suite of variables
describing defensible space for burned and unburned structures
within the perimeters of major fires from 2001 to 2010 in San
Diego County to ask the following questions:
1. How much defensible space is needed to provide significant
protection to homes during wildfires, and is it beneficial to
have more than the legally required 30 m (100 ft)?
2. Does the amount of defensible space needed for protection
depend on slope inclination?
3. What is the role of defensible space relative to other factors
that influence structure loss, such as terrain, fuel type and
housing density?
Methods
Study area
The properties and structures analysed were located in San
Diego County, California, USA (Fig. 1) – a topographically
diverse region with a Mediterranean climate characterised by
cool, wet winters and long summer droughts. Fire typically is a
direct threat to structures adjacent to wildland areas. Native
shrublands in southern California are extremely flammable
during the late summer and fall (autumn) and when ignited, burn
in high-intensity, stand-replacing crown fires. Although 500
homes on average have been lost annually since the mid-1900s
(Calfire 2000), that rate has doubled since 2000. Most of these
homes have burned during extreme fire weather conditions that
accompany the autumn Santa Ana winds. The wildland–urban
interface here includes more than 5 million homes, covering
more than 28 000 km
2
(Hammer et al. 2007).
Property data
The data for properties to analyse came from a complete spatial
database of existing residential structures and their
BInt. J. Wildland Fire A. D. Syphard et al.

corresponding property boundaries developed for San Diego
County (Syphard et al. 2012). This dataset included 687 869
structures, of which 4315 were completely destroyed by one of
40 major fires that occurred from 2001 to 2010. Our goal was to
compare homes that were exposed to wildfire and survived with
those that were exposed and destroyed. To determine exposure
to fire, we only considered structures located both within a GIS
layer of fire perimeters and within areas mapped as having
burned at a minimum of low severity through thematic Moni-
toring Trends in Burn Severity produced by the USA Geological
Survey and USDA Forest Service. From these data, we used a
random sample algorithm in GIS software to select 1000
destroyed and 1000 unburned homes that were not adjacent to
each other, to minimise any potential for spatial autocorrelation.
Our final property dataset included structures that burned across
eight different fires. More than 97% of these structures burned in
Santa Ana wind-driven fire events (Fig. 1).
Calculating defensible space and additional explanatory
variables
To estimate defensible space, we developed and explored a suite
of variables relative to the distance and amount of defensible
space surrounding structures, as well as the proximity of woody
vegetation to the structure (Table 1). We measured these vari-
ables based on interpretation of Google Earth aerial imagery.
We based our measurements on the most recent imagery before
the date of the fire. In almost all cases, imagery was available for
less than 1 year before the fire.
Our definition of defensible space followed the guidelines
published by the California Department of Forestry and Fire
Protection (Calfire 2006). ‘Clearance’ included all areas that
were not covered by woody vegetation, including paved areas
or grass. Although Google Earth prevents the identification of
understorey vegetation, woody trees and shrubs were easily
distinguished from grass, and our objective was to measure
horizontal distances as required by Calfire rather than assess the
relative flammability of different vegetation types. Trees or
shrubs were allowed to be within the defensible space zone as
long as they were separated by the minimum horizontal required
distance, which was 3 m (10 ft) from the edge of one tree canopy
to the edge of the next (Fig. 2). Although greater distances
between trees or shrubs are recommended on steeper slopes, we
followed the same guidelines for all properties. For all struc-
tures, we started the distance measurements by drawing lines
from the centre of the four orthogonal sides of the structure that
ended when they intersected anything that no longer met the
requirements in the guidelines. A fair number of structures are
not four sided; thus, the start of the centre point was placed at a
location that approximated the farthest extent of the structure
along each of four orthogonal sides.
We developed two sets of measurements of the distance of
defensible space based on what is feasible for homeowners
within their properties v. the total effective distance of defensi-
ble space. We made these two measurements because home-
owners are only required to create defensible space within their
own property, and this would reflect the effect of individual
homeowner compliance. Therefore, even if cleared vegetation
extended beyond the property line, the first set of distance
measurements ended at the property boundary. The second set
of measurements ignored the property boundaries and
accounted for the total potential effect of treatment. For all
measurements, we recorded the cover types (e.g. structure .3m
(10 ft) long, property boundary, or vegetation type) at which the
distance measurements stopped (Table 1). Because property
Destroyed
Unburned
N
Nevada
California
Fig. 1. Location of destroyed and unburned structures within the South Coast ecoregion of San Diego County, California, USA.
Defensible space for structure protection Int. J. Wildland Fire C

owners usually can only clear vegetation on their own land, it is
possible that the effectiveness of defensible space partly
depends upon the actions of neighbouring homeowners.
Therefore, we also recorded whether or not any neighbours’
un-cleared vegetation was located within 30 m (100 ft) of the
structure.
To assess the total amount of woody vegetation that can
safely remain on a property and still receive significant benefits
of defensible space, we calculated the total percentage of cleared
land, woody vegetation and structure area across every property.
This was accomplished by overlaying a grid on each property
and determining the proportion of squares falling into each class.
Preliminary results showed these three measurements to be
highly correlated, so we only retained percentage clearance
for further analysis. To evaluate the relative effect of woody
vegetation directly adjacent to structures, we also calculated the
number of sides of the structure with vegetation touching and
recorded whether any trees were overhanging structures’ roofs.
In addition to defensible space measurements, we evaluated
other factors known to influence the likelihood of housing loss to
fire in the region (Syphard et al. 2012,2013). Using the same
data as in Syphard et al. (2012,2013), we extracted spatial
information from continuous grids of explanatory variables for
the locations of all structures in our analysis. Variables included
interpolated housing density based on a 1-km search radius;
percentage slope derived from a 30-m digital elevation model
(DEM); Euclidean distance to nearest major and minor road and
fuel type, which was based on a simple classification of US
Forest Service data (Syphard et al. 2012), including urban, grass,
shrubland and forest & woodland.
1 – Urban veg
1
4
Residential
structure
Residential
structure
10 ft
Out-of-compliance
urban vegetation
In-compliance urban
vegetation
Wildland vegetation
Grass or bare ground
Total distance
defensible space
Property boundary
Legend
Distance defensible
space within property
3
2
2 – Urban to wildland
3 – Wildland veg
4 – Structure
Residential
structure
Fig. 2. Illustration of defensible space measurements. See Table 1 for full definition of terms.
Table 1. Defensible space variables measured for every structure
Urban veg, landscaping vegetation that was not in compliance with regulations within urban matrix; wildland veg, wildland vegetation that was not in
compliance with regulations; orchard, shrub to tree-sized vegetation in rows; urban to wildland, landscaping vegetation that leads into wildland vegetation;
structure, any building longer than 3 m (10 ft)
Variable Definition
Distance defensible space within property Measure of clearance from side of structure to property boundary calculated for four orthogonal directions
from structure and averaged
Total distance defensible space Measure of clearance from side of structure to end of clearance calculated for four orthogonal directions
from structure and averaged
Cover type at end of defensible space Type of cover encountered at end of measurement (urban veg, wildland veg, orchard, urban to wildland,
structure)
Percentage clearance Percentage of clearance calculated across the entire property
Neighbours’ vegetation Binary indicator of whether neighbours’ uncleared vegetation was located within 30 m (100 ft) of the main
structure
Vegetation touching structure Number of sides on which woody vegetation touches main structure (1–4) Structure with more than 4 sides
were viewed as a box and given a number between 1 and 4
Vegetation overhanging roof Was vegetation overhanging the roof? (yes or no)
DInt. J. Wildland Fire A. D. Syphard et al.

Analysis
We performed several analyses to determine whether relative
differences in home protection are provided by different dis-
tances and amounts of defensible space, particularly beyond
the legally required 30 m (100 ft), and to identify the effective
treatment distance for homes on low and steep slopes.
Categorical analysis
For the first analysis, we divided our data into several groups to
identify potential differences among specific categories of
defensible space distance around structures located on shallow
and steep slopes. We first sorted the full dataset of 2000 struc-
tures by slope and then split the data in the middle to create
groups of homes with shallow slope and steep slope. We divided
the data in half to keep the number of structures even within both
groups and to avoid specifying an arbitrary number to define
what constitutes shallow or steep slope. The two equal-sized
subsets of data ranged from 0 to 9%, with a mean of 8% for
shallow slope, and from 9 to 40%, with a mean of 27% for
steep slope. Within these data subsets, we next created groups
reflecting different mean distances of defensible space around
structures. We also performed separate analyses based on
whether defensible space measurements were calculated within
the property boundary or whether measurements accounted for
the total distance of defensible space.
Within all groups, we calculated the proportion of homes that
were destroyed by wildfire. We performed Pearson’s Chi-square
tests of independence to determine whether or not the proportion
of destroyed structures within groups was significantly different
(Agresti 2007). We based one test on four equal-interval groups
within the legally required distance of 30 m (100 ft): 0–7 m
(0–25 ft), 8–15 m (26–50 ft), 16–23 m (51–75 ft) and 24–30 m
(76–100 ft). A second test was based on three groups (24–30 m
(75–100 ft), 31–90 m (101–300 ft) and .90 m (.300 ft) or
.60 m (.200 ft)) to evaluate whether groups with mean defensi-
ble space distances .30 m (.100 ft) were significantly different
from groups with ,30 m (,100 ft). When defensible space
distances were only measured to the property boundary, few
structures had mean defensible space .90 m (.300 ft). Therefore,
we used a cut-off of 60 m (200 ft) to increase the sample size in
the Chi-square analysis. In addition to the Chi-square analysis, we
calculated the relative risk among every successive pair of
categories (Sheskin 2004). The relative risk was calculated as
the ratio of proportions of burned homes within two groups of
homes that had different defensible space distances.
Effective treatment analysis
In addition to comparing the relative effect of defensible space
among different groups of mean distances, as described above,
we also considered that the protective effect of defensible space
for structures exposed to wildfire is conceptually similar to the
effect of medication in producing a therapeutic response in
people who are sick. In addition to pharmacological applica-
tions, treatment–response relationships have been used for
radiation, herbicide, drought tolerance and ecotoxicological
studies (e.g. Streibig et al. 1993;Cedergreen et al. 2005;
Knezevic et al. 2007;Kursar et al. 2009). The effect produced
by a drug or treatment typically varies according to the
concentration or amount, often up to a point at which further
increase provides no additional response. The effective treat-
ment (ET50), therefore, is a specific concentration or exposure
that produces a therapeutic response or desired effect. Here we
considered the treatment to be the distance or amount of
defensible space.
Using the software package DRC in R (Knezevic et al. 2007;
Ritz and Streibig 2013), we evaluated the treatment–response
relationship of defensible space in survival of structures during
wildfire. To calculate the effective treatment, we fit a log-
logistic model with logistic regression because we had a binary
dependent variable (burned or unburned). We specified a
2-parameter model where the lower limit was fixed at 0 and
the upper limit was fixed at 1. We again performed separate
analyses for data subsets reflecting shallow and steep slope, as
well as from measurements of defensible space taken within, or
regardless of, property boundaries. We also performed analyses
to find the effective treatment of percentage clearance of trees
and shrubs within the property.
Multiple regression analysis
To evaluate the role of defensible space relative to other vari-
ables, we developed multiple generalised linear regression
models (GLMs) (Venables and Ripley 1994). We again had a
binary dependent variable (burned versus unburned), so we
specified a logit link and binomial response. Although the pro-
portion of 0s and 1s in the response may be important to consider
for true prediction (King and Zeng 2001;Syphard et al. 2008),
our objective here was solely to evaluate variable importance.
We developed multiple regression models for all possible
combinations of the predictor variables and used the corrected
Akaike’s Information Criterion (AICc) to rank models and
select the best ones for each region using package MuMIn in R
(R Development Core Team 2012;Burnham and Anderson
2002). We recorded all top-ranked models that had an AICc
value within 2 of that of the model with lowest AICc to identify
all models with empirical support. To assess variable impor-
tance, we calculated the sum of Akaike weights for all models
that contained each variable. On a scale of 0–1, this metric
represents the weight of evidence that models containing the
variable in question are the best model (Burnham and Anderson
2002). The distance of defensible space measured within
property boundaries was highly correlated with the distance of
defensible space measured beyond property boundaries
(r¼0.82), so we developed two separate analyses – one using
variables measured only within the property boundary and the
other using variables that accounted for defensible space outside
of the property boundary as well as the potential effect of
neighbours having uncleared vegetation within 30 m (100 ft) of
the structure. A test to avoid multicollinearity showed all other
variables within each multiple regression analysis to be uncor-
related (r,0.5).
Surrounding matrix
To assess whether the proportion of destroyed structures varied
according to their surrounding matrix, we summarised the most
common cover type at the end of defensible space measurements
(descriptions in Table 1) for all structures. These summaries
Defensible space for structure protection Int. J. Wildland Fire E

were based on the majority surrounding cover type from the four
orthogonal sides of the structure. We also noted cases in which
there was a tie (e.g. two sides were urban vegetation and two
sides were structures).
Results
Categorical analysis
When the distance of defensible space was measured both ‘only
within property boundaries’ (Fig. 3) and ‘regardless of property
boundaries’ (Fig. 4), the Chi-square test showed a significant
difference (P,0.001) in the proportion of destroyed structures
among the four equal-interval groups of distance ranging from
0 to 30 m (0–100 ft). This relationship was consistent on both
shallow-slope and steep-slope properties, although the relative
risk analysis showed considerable variation among classes
(Table 2) There was a steadily decreasing proportion of
destroyed structures at greater distances of defensible space up
to 30 m (100 ft) on the steep-slope structures with defensible
space measured regardless of property boundaries (Fig. 4b).
Otherwise, the biggest difference in proportion of destroyed
structures occurred between 0 and 7 m (0–25 ft) and 8–15 m
(26–50 ft) (Figs 3a–b,4a).
When the distance of defensible space was measured in
intervals from 24 m (75 ft) and beyond, the Chi-square test
showed no significant difference among groups (P¼0.96 for
shallow-slope properties and P¼0.74 for steep-slope proper-
ties) (Figs 3,4), although again, the relative risk analysis
showed considerable variation (Table 2).There was a slight
increase in the proportion of homes destroyed at longer distance
intervals when the defensible space was measured only to the
property boundaries (Fig. 3a–b). This slight increase is less
apparent when distances were measured regardless of bound-
aries (Fig. 4a–b).
The relative risk calculations showed that the ratio of
proportions was generally more variable among successive
pairs when the distances were measured within property
boundaries (Table 2). For these calculations, the risk of a
structure being destroyed was significantly lower when the
defensible space distance was 8–15 m (25–50 ft) compared
to 0–7 m (0–25 ft) on both shallow- and steep-slope properties.
On the steep-slope properties, there was an additional reduction
of risk when comparing 24–30 m (75–100 ft) to 16–23 m
(50–75 ft). However, the risk of a home being destroyed
was slightly significantly higher when there was 31–90 m
(101–225 ft) compared to 16–23 m (50–75 ft). For distances
that were measured regardless of property boundary (total
clearance), the only significant differences in risk of burning
were a reduction in risk for 8–15 m (25–50 ft) compared to
0–7 m (0–25 ft).
(a) Low slope properties
(b) High slope properties
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0–7 m 8–15 m 16–23 m 24–30 m 31–90 m 90⫹m
Proportion of homes destroyed
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0–7 m 8–15 m 16–23 m 24–30 m 31–90 m 90⫹m
Fig. 3. Proportion of destroyed homes grouped by distances of defensible
space based upon total distance of clearance within property boundary, for
structures on (a) shallow slopes (mean 8%) and (b) steep slopes (mean 27%).
(a)
Low slope properties
(b)
High slope properties
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Proportion of homes destroyed
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0–7 m 8–15m 16–23 m 24–30 m 31–90 m 90⫹m
0–7 m 8–15 m 16–23m 24–30 m 31–90 m 90⫹m
Fig. 4. Proportion of destroyed homes grouped by distances of defensible
space based upon total distance of clearance regardless of property bound-
ary, for structures on (a) shallow slopes (mean 8%) and (b) steep slopes
(mean 27%).
FInt. J. Wildland Fire A. D. Syphard et al.

Effective treatment analysis
Analysis of the treatment–response relationships among defen-
sible space and structures that survived wildfire showed that,
when all structures are considered together, the mean actual
defensible space that existed around structures before the fires
was longer than the calculated effective treatment (Table 3).
Regardless of whether the defensible space was measured within
or beyond property boundaries, the estimated effective treatment
of defensible space was nearly the same at 10 m (32–33 ft).
The effective treatment distance was much shorter for struc-
tures on shallow slopes (4–5 m (13–16 ft)) than for structures on
steep slopes (20–25 m (65–82 ft)), but in all cases was ,30 m
(,100 ft). Although longer distances of defensible space were
calculated as effective on steeper slopes, these structures actually
had shorter mean distances of defensible space around their
properties than structures on low slopes (Table 3).
The calculated effective treatment of the mean percentage
clearance on properties was 36% for all properties, 31% for
structures on shallow slopes and 37% for structures on steep
slopes (Table 3). In total, the properties all had higher actual
percentage clearance on their property than was calculated
to be effective. However, this mainly reflects the shallow-slope
properties, as those structures on steep slopes had less clearance
than the effective treatment.
Multiple regression analysis
When defensible space was measured only to the property
boundaries, it was not included in the best model, according to
the all-subsets multiple regression analysis (Table 4). However,
it was included in the best model when factoring in the distance
of defensible space measured beyond property boundaries
(Table 5). In both multiple regression analyses, low housing
density and shorter distances to major roads were ranked as the
most important variables according to their Akaike weights.
Slope and surrounding fuel type were also in both of the best
models as well as other measures of defensible space, including
the percentage clearance on property and whether vegetation
was overhanging the structure’s roof. The number of sides in
which vegetation was touching the structure was included in the
best model when defensible space was only measured to the
property boundary. The total explained deviance for the multi-
ple regression models was low (12–13%) for both analyses.
Table 2. Number of burned and unburned structures within defensible space distance categories (m), their relative risk and significance
A relative risk of 1 indicates no difference; ,1 means the chance of a structure burning is less than the other group; .1 means the chance is higher than the other
group. The relative risk is calculated for pairs that include the existing row and the row above. Confidence intervals are in parentheses
Distance within property Total distance
Burned Unburned Relative risk PBurned Unburned Relative risk P
Shallow slope
0–7 200 186 162 114
8–15 109 198 0.69 (0.12) ,0.001 108 132 0.77 0.002
16–23 51 89 1.03 (0.30) 0.850 78 90 1.03 0.770
24–30 36 40 1.30 (0.39) 0.110 50 70 0.90 0.430
31–90 28 47 0.79 (0.24) 0.220 79 99 1.06 0.640
60 or 90þ10 6 1.67 (0.63) 0.040 8 9 1.01 0.830
Steep slope
0–7 245 128 224 128
8–15 174 148 0.82 (0.10) 0.001 158 139 0.84 0.008
16–23 85 68 1.03 (0.16) 0.750 73 83 0.87 0.210
24–30 29 56 0.61 (0.17) 0.004 26 50 0.73 0.080
31– 29 28 1.49 (0.48) 0.050 39 68 1.06 0.760
60 or 90þ5 5 0.98 (0.47) 0.950 4 8 0.91 0.830
Table 3. Effective treatment results reflecting the distance (in metres, with feet in parentheses) and percentage clearance within properties that
provided significant improvement in structure survival during wildfires
The property mean is the average distance of defensible space or percentage clearance that was calculated on the properties before the wildfires and provides
a means to compare the effective treatment result to the actual amount on the properties
All parcels
effective
treatment
(n¼2000)
Parcel
mean
Shallow slope
(mean 8%)
effective treatment
(n¼1000)
Parcel
mean
Steep slope
(mean 27%)
effective treatment
(n¼1000)
Parcel
mean
Defensible space within parcel 10 (33) 13 (44) 4 (13) 14 (45) 25 (82) 11 (35)
Total distance defensible space 10 (32) 19 (63) 5 (16) 20 (67) 20 (65) 18 (58)
Mean percentage clearance on property 36 48 31 51 37 35
Defensible space for structure protection Int. J. Wildland Fire G

Surrounding matrix
The cover type that most frequently surrounded the structures at
the end of the defensible space measurements was urban vege-
tation, followed by urban vegetation leading into wildland
vegetation, and wildland vegetation (Fig. 5). Many structures
were equally surrounded by different cover types. There were no
significant differences in the proportion of structures destroyed
depending on the surrounding cover type. However, a dispro-
portionately large proportion of structures burned (28 v. 9%
unburned) when they were surrounded by urban vegetation that
extended straight into wildland vegetation.
Discussion
For homes that burned in southern Californian urban areas
adjacent to non-forested ecosystems, most burned in high-
intensity Santa Ana wind-driven wildfires and defensible space
increased the likelihood of structure survival during wildfire.
The most effective treatment distance varied between 5 and
20 m (16–58 ft), depending on slope and how the defensible
space was measured, but distances longer than 30 m (100 ft)
provided no significant additional benefit. Structures on steeper
slopes benefited from more defensible space than structures on
shallow slopes, but the effective treatment was still less than
30 m (100 ft). The steepest overall decline in destroyed struc-
tures occurred when mean defensible space increased from
0–7 m (0–25 ft) to 8–15 m (26–50 ft). That, along with the
multiple regression results showing the significance of vegeta-
tion touching or overhanging the structure, suggests it is most
critical to modify vegetation immediately adjacent to the house,
and to move outward from there. Similarly, vegetation over-
hanging the structure was also strongly correlated with structure
loss in Australia (Leonard et al. 2009).
In terms of fuel modification, the multiple regression models
also showed that the percentage of clearance was just as, or
more important than, the linear distance of defensible space.
Table 4. Results of multiple regression models of destroyed homes using all possible variable combinations and
corrected Akaike’s Information Criterion (AICc)
Includes variables measured within property boundary only. Top-ranked models include all those (n¼12) with AICc within 2 of
the model with the lowest AICc. Relative variable importance is the sum of ‘Akaike weights’ over all models including the
explanatory variable
Variable in order of importance Relative variable
importance
Model-averaged
coefficient
Number inclusions in
top-ranked models
Housing density 1 0.003 12
Distance to major road 1 0.0005 12
Percentage clearance 1 0.02 12
Slope 1 0.03 12
Vegetation overhang roof 1 0.5 12
Fuel type 0.67 Factor 9
Vegetation touch structure 0.49 0.07 6
Distance defensible space within property 0.45 0.0002 5
South-westness 0.36 0.0007 3
Distance to minor road 0.28 0.0002 1
D
2
of top-ranked model 0.123
Table 5. Results of multiple regression models of destroyed homes using all possible variable combinations and corrected
Akaike’s Information Criterion (AICc)
Includes variables measured beyond property boundary. Top-ranked models include all those (n¼6) with AICc within 2 of the model
with the lowest AICc. Relative variable importance is the sum of ‘Akaike weights’ over all models including the explanatory variable
Variable in order of importance Relative variable
importance
Model-averaged
coefficient
Number inclusions in
top-ranked models
Housing density 1 0.003 6
Distance to major road 1 0.0005 6
Total distance defensible space 1 0.004 6
Percentage clearance 1 0.01 6
Vegetation overhang roof 0.99 0.4 6
Slope 0.99 0.03 6
Fuel type 0.86 Factor 4
South-westness 0.42 0.0009 2
Distance to minor road 0.36 0.0009 2
Neighbours’ vegetation 0.27 0.08 1
Vegetation touch structure 0.27 0.18 1
D
2
of top-ranked model 0.125
HInt. J. Wildland Fire A. D. Syphard et al.

However, as with defensible space, percentage clearance did not
need to be draconian to be effective. Even on steep slopes, the
effective percentage clearance needed on the property was
,40%, with no significant advantage beyond that. Although
these steep-slope structures benefited more from clearance, they
tended to have less clearance than the effective amount, which
may be why slope was such an important variable in the multiple
regression models. Shallow-slope structures, in contrast, had
more clearance on average than was calculated to be effective,
suggesting these property owners do not need to modify their
behaviours as much relative to people living on steep slopes.
Although the term ‘clearance’ is often used interchangeably
with defensible space, this term is incorrect when misinterpreted
to mean clearing all vegetation, and our results underline this
difference. The idea behind defensible space is to reduce the
continuity of fuels through maintenance of certain distances
among trees and shrubs. Although we could not identify the
vertical profile of fuels through Google Earth imagery, the fact
that at least 60% of the horizontal woody vegetative cover can
remain on the property with significant protective effects
demonstrates the importance of distinguishing defensible space
from complete vegetation removal. Thus, we suggest the term
‘clearance’ be replaced with ‘fuel treatment’ as a better way of
communicating fire hazard reduction needs to home owners.
The percentage cover of woody shrubs and trees was not
evenly distributed across properties, and we did not collect data
describing how the cover was distributed. Considering the
importance of defensible space and vegetation modification
immediately adjacent to the structure, it should follow that
actions to reduce cover should also be focussed in close
proximity to the structure. The hazard of vegetation near the
structure has apparently been recognised for some time (Foote
et al. 1991;Ramsey and McArthur 1994), but it is not stressed
enough, and rarely falls within the scope of defensible space
guidelines or ordinances.
In addition to the importance of vegetation overhanging or
touching the structure, it is important to understand that orna-
mental vegetation may be just as, if not more, dangerous than
native vegetation in southern California. Although the results
showed no significant differences in the cover types in the
surrounding matrix, there was a disproportionately large number
of structures destroyed (28% burned v. 9% unburned) when
ornamental vegetation on the property led directly into the
wildland. Ornamental vegetation may produce highly flamma-
ble litter (Ganteaume et al. 2013) or may be particularly
dangerous after a drought when it is dry, or has not been
maintained, and species of conifer, juniper, cypress, eucalypt,
Acacia and palm have been present in the properties of many
structures that have been destroyed (Franklin 1996). Neverthe-
less, ornamental vegetation is allowed to be included as defen-
sible space in many codes and ordinances (Haines et al. 2008).
One reason that longer defensible space distances did not
significantly increase structure protection may be that most
homes are not destroyed by the direct ignition of the fire front
but rather due to ember-ignited spot fires, sometimes from fire
brands carried as far as several km away. Although embers
decay with distance, the difference between 30 and 90 m (100
and 300 ft) may be small relative to the distance embers travel
under the severe wind conditions that were present at the time of
the fires. The ignitability of whatever the embers land on,
particularly adjacent to the house, is therefore most critical for
propagating the fire within the property or igniting the home
(Cohen 1999;Maranghides and Mell 2009).
Aside from roofing or home construction materials and
vegetation immediately adjacent to structures (Quarles et al.
2010;Keeley et al. 2013), the flammability of the vegetation in
the property may also play a role. Large, cleared swaths of land
are likely occupied at least in part by exotic annual grasses that
are highly ignitable for much of the year. Conversion of woody
shrubs with higher moisture content into low-fuel-volume grass-
lands could potentially increase fire risk in some situations by
increasing the ignitability of the fuel; and if the vegetation
between a structure and a fire is not readily combustible, it could
protect the structure by absorbing heat flux and filtering fire
brands (Wilson and Ferguson 1986).
The slight increase in proportion of structures destroyed with
longer distances of defensible space within parcel boundaries
was surprising. However, that increase was not significant in the
Chi-square analysis, although there were some significant
differences in the pairwise relative risk analysis. Nevertheless,
the largest significant effect of defensible space was between the
categories of 0–7 m (0–25 ft) to 8–15 m (26–50 ft), and it may be
that differences in categories beyond these distances are not
highly meaningful or reflect an artefact of the definition of
distance categories. These relationships at longer distances are
likely also weak compared to the effect of other variables
operating at a landscape scale. Although the categorical analysis
allowed us to answer questions relative to legal requirements
and specific distances, the effective treatment analysis was
important for identifying thresholds in the continuous variable.
The multiple regression models showed that landscape
factors such as low housing density and longer distances to
major roads were more important than distance of defensible
space for explaining structure destruction, and the importance of
0
0.2
0.4
0.6
0.8
1.0
Wildland
ve
g
etation
Urban
ve
g
etation
Urban to
wildland
Orchard Structure Tie
Proportion of structures
Burned
Unburned
Fig. 5. Proportion of destroyed and unburned structures based on the
primary surrounding cover type at the end of defensible space measure-
ments. There were no significant differences in the proportion of burned and
unburned structures within cover types (P¼0.14). Cover types are defined
in Table 1.
Defensible space for structure protection Int. J. Wildland Fire I

these variables is consistent with previous studies (Syphard et al.
2012,2013), despite the smaller spatial extent studied here.
Whereas this study used an unburned control group exposed to
the same fires as the destroyed structures, previous studies
accounted for structures across entire landscapes. The likeli-
hood of a fire destroying a home is actually a result of two major
components: the first is the likelihood that there will be a fire,
and the second is the likelihood that a structure will burn in that
fire. In this study, we only focussed on structure loss given the
presence of a fire, and the total explained variation for the
multiple regression models was quite low at ,12%. However,
when the entire landscape was accounted for in the total
likelihood of structure destruction, the explained variation of
housing density alone was .30% (Syphard et al. 2012). One
reason for the relationship between low housing density and
structure destruction is that structures are embedded within a
matrix of wildland fuel that leads to greater overall exposure,
which is consistent with Australian research that showed a linear
decrease of structure loss with increased distance to forest (Chen
and McAneney 2004). That research, however, only focussed on
distance to wildland boundaries and did not quantify variability
in defensible space or ornamental vegetation immediately
surrounding structures. Thus, fire safety is important to consider
at multiple scales and for multiple variables, which will ulti-
mately require the cooperation of multiple stakeholders.
Conclusions
Structure loss to wildfire is clearly a complicated function of
many biophysical, human and spatial factors (Keeley et al.
2009;Syphard et al. 2012). For such a large sample size, we
were unable to account for home construction materials, but this
is also well understood to be a major factor, with older homes
and wooden roofs being most vulnerable (Franklin 1996;Cohen
1999,2000). In terms of actionable measures to reduce fire risk,
this study shows a clear role for defensible space up to 30 m
(100 ft). Although the effective distances were on average much
shorter than 30 m (100 ft), we recognise that additional distance
may be necessary to provide sufficient protection to firefighters,
which we did not address in this study (Cheney et al. 2001). In
contrast, the data in this study do not support defensible space
beyond 30 m (100 ft), even for structures on steep slopes. In
addition to the fact that longer distances did not contribute
significant additional benefit, excessive vegetation clearance
presents a clear detriment to natural habitat and ecological
resources. Results here suggest the best actions a homeowner
can take are to reduce percentage cover up to 40% immediately
adjacent to the structure and to ensure that vegetation does not
overhang or touch the structure.
In addition to defensible space, this study also underlines the
potential importance of land use planning to develop communi-
ties that are fire safe in the long term, in particular through their
reduction to exposure to wildfire in the first place. Localised
subdivision decisions emphasising infill-type development pat-
terns may significantly reduce fire risk in the future, in addition
to minimising habitat loss and fragmentation (Syphard et al.
2013). This study was conducted in southern California, which
has some of the worst fire weather in the world and many
properties surrounded by large, flammable exotic trees.
Therefore, recommendations here should apply to other non-
forested ecosystems as well as many forested regions.
Acknowledgements
We acknowledge funding from the US Geological Survey Fire Risk Scenario
Project and note that use of trade, product, or firm names is for descriptive
purposes only and does not imply endorsement by the US Government.
References
Absher JD, Vaske JJ (2011) The role of trust in residents’ fire wise actions.
International Journal of Wildland Fire 20, 318–325. doi:10.1071/
WF09049
Agresti A (2007) ‘An Introduction to Categorical Data Analysis’, 2nd edn.
(Wiley: New York)
Blanchi R, Lucas C, Leonard J, Finkele K (2010) Meteorological condi-
tions and wildfire-related house loss in Australia. International Journal
of Wildland Fire 19, 914–926. doi:10.1071/WF08175
Boschetti L, Roy D, Barbosa P, Justice C (2008) A MODIS assessment of
the summer 2007 extent burned in Greece. International Journal of
Remote Sensing 29, 2433–2436. doi:10.1080/01431160701874561
Burnham KP, Anderson DR (2002) ‘Model Selection and Multimodel
Inference: a Practical Information-Theoretic Approach’, 2nd edn.
(Springer-Verlag: New York)
Calfire (2000) ‘Wildland Fire Hazard Assessment Final Report on
FEMA.’ (California Department of Forestry and Fire Protection:
Sacramento, CA)
Calfire (2006) ‘General Guidelines for Creating Defensible Space.’
(California Department of Forestry and Fire Protection: Sacramento, CA)
Cary GJ, Flannigan MD, Keane RE, Bradstock RA, Davies ID, Lenihan
JM, Li C, Logan KA, Parsons RA (2009) Relative importance of fuel
management, ignition management and weather for area burned: evi-
dence from five landscape–fire–succession models. International Jour-
nal of Wildland Fire 18, 147–156. doi:10.1071/WF07085
Cedergreen N, Ritz C, Streibig JC (2005) Improved empirical models for
describing hormesis Environmental Toxicology and Chemistry 24,
3166–3177. doi:10.1897/05-014R.1
Chen K, McAneney J (2004) Quantifying bushfire penetration into urban
areas in Australia. Geophysical Research Letters 31, L12212.
doi:10.1029/2004GL020244
Cheney P, Gould J, McCaw L (2001) The dead-man zone – a neglected area
of firefighter safety. Australian Forestry 64, 45–50. doi:10.1080/
00049158.2001.10676160
Cohen JD (1999) Reducing the wildland fire threat to homes: where and how
much? In ‘Proceedings of the Symposium on Fire Economics, Planning,
and Policy: Bottom Lines’, 5–9 April 1999, San Diego, CA. (Eds
A Gonzales-Caban, PN Omi) USDA Forest Service, Pacific Southwest
Research Station, General Technical Report PSW-GTR-173, pp. 189–195.
(Albany, CA)
Cohen JD (2000) Home ignitability in the wildland–urban interface. Journal
of Forestry 98, 15–21.
Filmon G (2004) Firestorm 2003, provincial review. Report to the Provincial
Government of British Columbia (Vancouver, BC) Available at http://
bcwildfire.ca/History/ReportsandReviews/2003/FirestormReport.pdf
[Verified 9 August 2014]
Foote EID, Gilless JK (1996) Structural survival. In ‘California’s I-Zone.’
(Ed R Slaughter) pp. 112–121. (CFESTES: Sacramento, CA)
Foote EID, Martin RE, Gilless JK (1991) The defensible space factor study:
a survey instrument for post-fire structure loss analysis. In ‘Proceedings
of the 11th Conference on Fire and Forest Meteorology’, 16 April 1991,
Bethesda, MD. (Eds PL Andrews, DF Potts) pp. 66–73. (Society of
American Foresters: Bethesda, MD)
Franklin SE (1996) California’s catastrophic intermix fires causes, culprits
and cures. American Fire Journal 48, 20–23.
JInt. J. Wildland Fire A. D. Syphard et al.

Ganteaume A, Jappiot M, Corrine L (2013) Assessing the flammability of
surface fuels beneath ornamental vegetation in wildland–urban inter-
faces in Provence (south-eastern France). International Journal of
Wildland Fire 22, 333–342. doi:10.1071/WF12006
Gibbons P, van Bommel L, Gill MA, Cary GJ, Driscoll DA, Bradstock RA,
Knight E, Moritz MA, Stephens SL, Lindenmayer DB (2012) Land
management practices associated with house loss in wildfires. PLoS
ONE 7, e29212. doi:10.1371/JOURNAL.PONE.0029212
Gill AM, Stephens SL (2009) Scientific and social challenges for the
management of fire-prone wildland–urban interfaces. Environmental
Research Letters 4, 034014. doi:10.1088/1748-9326/4/3/034014
Gilmer M (1994) ‘California Wildfire Landscaping.’ (Taylor Publishing
Company: Dallas, TX)
Gude PH, Rasker R, van den Noort J (2008) Potential for future develop-
ment on fire-prone lands. Journal of Forestry 106, 198–205.
Haines TK, Renner CR, Reams MA (2008) A review of state and local
regulation for wildfire mitigation. In ‘The Economics of Forest Dis-
turbances: Wildfires, Storms, and Invasive Species’. (Eds TP Holmes,
JP Prestemon, KL Abt) pp. 273–293. (US Forest Service: Washington, DC)
Available at http://www.treesearch.fs.fed.us/pubs/32690 [Verified 9
August 2014]
Hammer RB, Radeloff VC, Fried JS, Stewart SI (2007) Wildland–urban
interface housing growth during the 1990s in California, Oregon, and
Washington. International Journal of Wildland Fire 16, 255–265.
doi:10.1071/WF05077
Hessl AE (2011) Pathways for climate change effects on fire: models,
data, and uncertainties Progress in Physical Geography 35, 393–407.
doi:10.1177/0309133311407654
Howard RA, North DW, Offensend FL, Smart CN (1973) ‘Decision
Analysis of Fire Protection Strategy for the Santa Monica Mountains:
an Initial Assessment.’ (Stanford Research Institute: Menlo Park, CA)
Keeley JE, Baer-Keeley M, Fotheringham CJ (2005) Alien plant dynamics
following fire in Mediterranean-climate California shrublands. Ecologi-
cal Applications 15, 2109–2125. doi:10.1890/04-1222
Keeley JE, Safford HD, Fotheringham CJ, Franklin J, Moritz MA (2009)
The 2007 southern California wildfires: lessons in complexity. Journal
of Forestry 107, 287–296.
Keeley JE, Syphard AD, Fotheringham CJ (2013) The 2003 and 2007
wildfires in southern California. In ‘Natural Disasters and Adaptation
to Climate Change’. (Eds S Boulter, J Palutikof, DJ Karoly, D Guitart)
pp. 42–52. (Cambridge University Press: Oxford, UK)
King G, Zeng L (2001) Logistic regression in rare events data. Political
Analysis 9, 137–163. doi:10.1093/OXFORDJOURNALS.PAN.A004868
Knezevic SZ, Streibig JC, Ritz C (2007) Utilizing R software package for
dose-response studies: the concept and data analysis. Weed Technology
21, 840–848. doi:10.1614/WT-06-161.1
Kursar TA, Engelbrecht BMJ, Burke A, Tyree MT, El Omari B, Giraldo JP
(2009) Tolerance to low leaf water status of tropical tree seedlings is
related to drought performance and distribution. Functional Ecology 23,
93–102. doi:10.1111/J.1365-2435.2008.01483.X
Leonard J, Blanchi R, Lipkin F, Newnham G, Siggins A, Opie K, Culvenor D,
CechetB, Corby N, Thomas C, HabiliN, Jakab M, Coghlan R, Lorenzin G,
Campbell D, Barwick M (2009) Building and land-use planning research
after the 7th February Victorian bushfires: preliminary findings. CSIRO
and Bushfire CRC. (Melbourne)
Maire RG (1979) ‘Landscape for Fire Protection.’ (University of California
Agriculture Extension Service: Los Angeles, CA)
Maranghides A, MellWE (2009) A case study of a community affected by the
Witch and Guejito fires. NIST Technical Note 1635. (Washington, DC)
Quarles SL, Valachovic Y, Nakamura GM, Nader GA, DeLasaux J (2010)
Home survival in wildfire-prone areas: building materials and design
considerations. University of California, Agriculture and Natural
Resources, ANR Publication 8393. (Richmond, CA) Available at
http://anrcatalog.ucdavis.edu/pdf/8393.pdf [Verified 9 August 2014]
R Development Core Team (2012) R: a language and environment for
statistical computing. (R Foundation for Statistical Computing, Vienna,
Austria) Available at http://www.R-project.org/ [Verified 23 August
2013]
Ramsey GC, McArthur NA (1994) Planning in fire-prone areas: building
survival. In ‘Bushfire! Looking to the Future: Papers from the Nature
Conservation Council of NSW Seminar’, June 1994. (Eds C Brown,
L Tohver) pp. 142–150. (Envirobook: Sydney)
Ritz C, Streibig JC (2013) Package ‘drc’ analysis of dose-response curves.
Available at http://cran.r-project.org/web/packages/drc/index.html[Veri-
fied 9 August 2014]
Schoennagel T, Nelson CR, Theobald DM, Carnwath GC, Chapman TB
(2009) Implementation of National Fire Plan treatments near the
wildland–urban interface in the western United States. Proceedings of
the National Academy of Sciences of the United States of America 106,
10 706–10 711. doi:10.1073/PNAS.0900991106
Sheskin DJ (2004) ‘Handbook of Parametric and Nonparametric Statistical
Procedures’, 3rd edn. (Chapman & Hall and CRC: Boca Raton, FL).
Smith E, Adams G (1991) ‘Incline Village/Crystal Bay Defensible Space
Handbook.’ (University of Nevada: Reno, NV)
Spittler TE (1995) Fire and the debris flow potential of winter storms. In
‘Brushfires in California Wildlands: Ecology and Resource Manage-
ment.’ (Eds JE Keeley, T Scott) pp. 113–120. (International Association
of Wildland Fire: Fairfield, WA)
Stockmann K, Burchfield J, Calkin D, Venn T (2010) Guiding preventative
wildland fire mitigation policy and decisions with an economic model-
ing system. Forest Policy and Economics 12, 147–154. doi:10.1016/
J.FORPOL.2009.09.009
Streibig JC, Rudemo M, Jensen JE (1993) Dose-response curves and
statistical models. In ‘Herbicide Bioassays.’ (Eds JC Streibig, P Kudsk)
pp. 29–55. (CRC: Boca Raton, FL)
Syphard AD, Franklin J, Keeley JE (2006) Simulating the effects of
frequent fire on southern California coastal shrublands. Ecological
Applications 16, 1744–1756. doi:10.1890/1051-0761(2006)016[1744:
STEOFF]2.0.CO;2
Syphard AD, Radeloff VC, Keuler NS, Taylor RS, Hawbaker TJ, Stewart
SI, Clayton MK (2008) Predicting spatial patterns of fire on a southern
California landscape. International Journal of Wildland Fire 17,
602–613. doi:10.1071/WF07087
Syphard AD, Keeley JE, Massada AB, Brennan TJ, Radeloff VC (2012)
Housing arrangement and location determine the likelihood of housing
loss due to wildfire. PLoS ONE 7, e33954. doi:10.1371/JOURNAL.
PONE.0033954
Syphard AD, Bar Massada A, Butsic V, Keeley JE (2013) Land use
planning and wildfire: development policies influence future probability
of housing loss. PLoS ONE 8, e71708. doi:10.1371/JOURNAL.PONE.
0071708
Vasquez T (2011) The Russian inferno of 2010. Weatherwise 64, 20–25.
doi:10.1080/00431672.2011.551592
Venables WM, Ripley BD (1994) ‘Modern Applied Statistics with S-Plus.’
(Springer-Verlag: New York)
Wilson AAG, Ferguson IS (1986) Predicting the probability of house
survival during bushfires. Journal of Environmental Management 3,
259–270.
Winter G, McCaffrey S, Vogt CA (2009) The role of community policies in
defensible space compliance. Forest Policy and Economics 11,
570–578. doi:10.1016/J.FORPOL.2009.07.004
Witter M, Taylor RS (2005) Preserving the future: a case study in fire
management and conservation from the Santa Monica Mountains. In
‘Fire, Chaparral, and Survival in Southern California’. (Ed. RW Halsey)
pp. 109–115. (Sunbelt Publications: San Diego, CA)
Za´ rate L, Arnaldos J, Casal J (2008) Establishing safety distances for
wildland fires. Fire Safety Journal 43, 565–575. doi:10.1016/J.FIRESAF.
2008.01.001
www.publish.csiro.au/journals/ijwf
Defensible space for structure protection Int. J. Wildland Fire K